Mitochondrial control of sleep

The function of sleep remains one of biology ’ s biggest mysteries. A solution to this problem is likely to come from a better understanding of sleep homeostasis, and in particular of the cellular and molecular processes that sense sleep need and settle sleep debt. Here, we highlight recent work in the fruit fly showing that changes in the mitochondrial redox state of sleep-promoting neurons lie at the heart of a homeostatic sleep-regulatory mechanism. Since the function of homeo-statically controlled behaviours is often linked to the regulated variable itself, these findings corroborate with the hypothesis that sleep serves a metabolic function.


Introduction
Sleep is a state of reversible quiescence characterized by a reduced responsiveness to sensory stimuli, the adoption of a species-specific posture, and physiological changes in brain activity [1]. Sleep evolved at least 600 million years ago and is highly conserved across the animal kingdom, including in vertebrates, nematodes, arthropods, and cnidarians [2]. The ubiquity of sleep and its presence in animals lacking a central nervous system, such as in the jellyfish of the Cassiopea genus [3], suggests that sleep is not only from the brain, for the brain [4] but is also critical for functions in the periphery [5].
Sleep is regulated by circadian and homeostatic processes [6,7]. While the circadian clock regulates the timing of sleep, the homeostat was proposed to track variables that accumulate during waking and induce sleep once a threshold has been surpassed. In the context of metabolism, the circadian clock and the sleep homeostat enforce the temporal segregation of mutually antagonistic metabolic processes into two states: wake and sleep [5,8,9]. For example, in fat tissue, lipid catabolism occurs during sleep, while lipid anabolism (lipogenesis) prevails during wake [10]. The temporal compartmentalization of metabolic processes in different tissues and cell types is critical for cellular function [11] and it is no surprise that clock perturbations and inadequate sleep (either too long or too short) are associated with metabolic disorders, such as diabetes [12], obesity [13] and cardiovascular disease [14]. Metabolism itself, and in particular the cellular redox state, also regulates sleep [15e17], suggesting a bidirectional relationship as well as shared neuronal circuits and molecular mechanisms. With mitochondria playing a central role in energy metabolism, the question arises as to what extent mitochondria play a role in regulating sleep and vice-versa. In the present opinion article, we discuss evidence surrounding the role of mitochondria in sleep regulation.

Mitochondria and sleep
Mitochondria play a pivotal role in energy metabolism by generating adenosine triphosphate (ATP) from fuel molecules through oxidative phosphorylation (OXPHOS) (BOX 1). In addition, they are involved in non-OXPHOS metabolic, as well as non-metabolic processes, such as fatty acid metabolism, apoptosis, and innate immunity [18]. Emerging evidence suggests that mitochondrial function extends beyond cellular and physiological processes and can influence an organism's memory [19,20], sociability [21,22], and sleep [16], thereby linking mitochondrial energetics, and specifically the cellular redox state, to behavioral states.
One of the main determinants of the cellular redox state is the level of reactive oxygen species (ROS). ROS is a collective term referring to superoxide, hydrogen peroxide, hydroxyl radical, and other reactive species derived from oxygen, some of which are highly reactive and can cause oxidative damage by attacking DNA, lipids, or proteins [23]. The production of mitochondrial ROS (mtROS) occurs during the electron transfer process of OXPHOS where electrons can leak and react with molecular oxygen (BOX 1). Since mtROS are obligate byproducts of mitochondrial aerobic respiration, the metabolic rate can determine the level of mtROS production (BOX 1).
The metabolic rate of an organism fluctuates with its sleep-wake rhythm: during sleep, metabolism is reduced whereas during wakefulness the metabolic rate is increased [24,25]. In support of this, there is an upregulation of genes related to energy metabolism during wake in comparison to sleep [26] and extended wakefulness is associated with increased energy expenditure [27e29]. Therefore, the rate of mtROS production is predicted to be higher during wakefulness due to an increase in OXPHOS. Because mitochondria are a major ROS production site and because ROS can be detrimental for cellular health, mitochondria could be prone to increasing levels of oxidative damage and stress as a function of waking time and/or experience. While several enzymatic and non-enzymatic antioxidant systems have evolved to neutralize ROS in-and outside mitochondria, exposure above a certain threshold is toxic and can pose a serious threat to cellular health. For this reason, sleep has been proposed to mitigate oxidative damage by removing free oxygen radicals that accumulate during wakefulness [30].
Several studies have tested this hypothesis by measuring changes in antioxidant enzyme activity, as well as markers of oxidative stress and/or damage upon sleep deprivation. Overall, the results obtained in brain tissue remain controversial: while some studies show indirect signs of oxidative stress by measuring antioxidant enzyme activity [31,32], they do not detect oxidative damage, while others report changes in oxidative damage, e.g., lipid peroxidation or DNA damage [33e35], or no changes at all [36e39]. Methodological aspects, however, often limit the comparison between these studies because of differences in, e.g., rodent strain, the brain regions analyzed, sample preparation, sleep deprivation protocols (including sleep-stage specific (i.e., rapid eye movement (REM) and non-rapid eye movement (NREM) sleep) vs. total sleep deprivation), experimental readouts, as well as the short-lived, reactive nature of ROS and the lack of appropriate probes. In addition, methodologies traditionally used to assess lipid peroxidation, such as the TBARS assay, are unspecific and may result in falsepositives [23]. The inconclusive evidence for widespread oxidative damage in the brain following extended wakefulness may, however, point towards the efficacy of the brain to protect itself from ROS-induced damage [26]. Indeed, neural antioxidant defenses are dynamically regulated by synaptic activity in newborn mice as well as in primary neuronal cultures, protecting the cell from activity-dependent increases in ROS [40]. It remains to be studied, however, whether this activity-dependent boosting of antioxidant defenses also occurs in vivo in the adult brain. Paradoxically, neurons in isolation have weak antioxidant defenses and heavily rely on nearby astrocytes instead, which possess high intracellular concentrations of antioxidants [20,41]. The cooperative role of astrocytes and neurons may provide the brain with an additional layer of protection from oxidative damage.
Recent findings suggest that organs other than the brain are prone to significant sleep deprivation-induced oxidative damage. Vaccaro and colleagues [39] found that acute and chronic sleep deprivation leads to ROS accumulation and oxidative stress in the gut rather than in the brain. This was accompanied by a shortening of lifespan that was rescued by targeting anti-oxidant enzymes in the gut [39]. This is in line with previous work BOX 1. Production of ROS in mitochondria.
Neurons rely on oxidative phosphorylation (OXPHOS) to produce energy in the form of adenosine triphosphate (ATP) from organic fuel molecules. During OXPHOS, electrons derived from the reducing equivalents nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FADH2) are transferred into the mitochondrial electron transport chain. This process is coupled to the pumping of protons from the matrix into the intermembrane space, which generates an electrochemical gradient. When protons are transported back into the matrix across the inner mitochondrial membrane, the resulting energy is, in turn, used to generate ATP, the currency of the cell. During the electron transfer process, electrons can leak and directly react with molecular oxygen to form superoxide (O 2 -) at complex I and III of the respiratory chain. Superoxide, in turn, can dismutate to hydrogen peroxide (H 2 O 2 ), which can react to form hydroxyl radical (HO . ) [45]. While O 2 .and H 2 O 2 have a limited capacity to attack biological molecules and can be neutralized by several cellular enzymatic anti-oxidants, HO . is highly reactive and indiscriminate in nature [23]. Within mitochondria, ROS can also be generated at multiple enzymatic sites other than respiratory complexes localized to the inner and outer mitochondrial membrane, such as a-glycerophosphate dehydrogenase (aGPDH) or monoamine oxidase (MAO), respectively [52]. While in some cases it can be directly produced at the enzyme per se, in others, it was shown to be generated within the electron transport chain due to electron transfer onto coenzyme Q [52]. Importantly, most of what we know about mitochondrial ROS species, levels and production sites comes from in vitro studies using isolated mitochondria or enzymes. This makes it difficult to extrapolate these observations to in vivo conditions, and to assess the physiological relevance of specific ROS species and production sites, as well as their contribution to a biological process [23]. Thus, the development of new probes, such as MitoNeoD [53], and genetically encoded tools will be critical to gain insights into ROS species identity, production sites and dynamics under physiological conditions in the future. Moreover, the multiplexed use of sensors for ATP [54], ATP/ADP [55], pH [56], NADH/NAD + [57] and ROS [58] will allow to determine whether different mitochondrial modes of operation also occur in vivo, and if so, whether and how they are modulated by the behavioral state of the animal and by the cellular identity.
from Everson and colleagues [42], who found significant oxidative DNA damage in the gut, liver, and lung of rats after sleep deprivation. This raises intriguing questions as to whether sleep promotes repair and homeostasis of peripheral tissues, in particular of the gut, and whether peripheral organs themselves participate in sleep induction by, e.g., broadcasting signals of cellular stress to sleep-regulatory regions in the brain. As such, ROS may not simply be a biproduct of sleep-wake dynamics, but may in itself be an integral signal for sleep regulation.

mtROS in homeostatic sleep control
Recent work in the fly revealed that mtROS acts as an instructive signal for homeostatic sleep regulation by regulating the excitability of a set of sleep-promoting neurons (Figure 1). In the fly brain, a specific cluster of neurons referred to as dorsal fan-shaped body (dFB) neurons change their excitability as a function of sleep need: the higher the sleep need of the fly, the more excitable the neurons, and vice versa [43]. This gave rise to the hypothesis that as sleep need builds up, dFB neurons upregulate their excitability, that is, their intrinsic ability to translate synaptic inputs into action potentials. Conversely, as sleep debt gets cleared, e.g., after rebound sleep, their excitability reverts back to baseline. In other words, sleep need modulates the input-output function of dFB neurons. Accordingly, genetic manipulations that prevent excitability changes in dFB neurons affect sleep and sleep homeostasis [43].
Until recently, it was unclear what mechanisms lead to changes in excitability but a surprising role for mtROS in the regulation of dFB excitability has now been uncovered. Prolonged periods of wakefulness were shown to be accompanied by a cumulative increase in mtROS levels, which is indirectly sensed at the plasma membrane by Hyperkinetic, a redox-sensitive potassium channel subunit. Hyperkinetic modulates the channel's inactivation kinetics and increases the excitability of dFB neurons thereby promoting sleep [16] (Figure 1). As a result, experimentally increasing ROS levels in dFB neurons is sufficient to promote sleep. Interestingly, similar sleep need-dependent changes in mtROS levels were absent in at least one other neuronal population previously implicated in sleep control. Therefore, sleep deprivation may act on a group of specialized cells dedicated to homeostatic sleep control, rather than cause brain-wide changes in oxidation. Together, these results suggest that a cell type-specific change in dFB mitochondrial oxidation state conveys sleep need via biophysical coupling of the neuron's metabolism with its excitability. To which extent dFB neurons monitor their own mitochondrial metabolism to simply gain an estimate of waking duration, and to which they act as Model for the build-up and reset of sleep need. In the fly brain, a cluster of neurons referred to as dorsal fan-shaped body (dFB) neurons are known to change their excitability as a function of sleep need: the more excitable the neurons, the higher the sleep need of the fly [43]. This gave rise to the following model: as sleep need builds up, dFB neurons upregulate their excitability, that is, their intrinsic ability to translate synaptic inputs into action potentials. Conversely, as sleep debt gets cleared, e.g., after rebound sleep, their excitability reverts back to baseline. In other words, sleep need modulates the input-output function of dFB neurons. A recent study showed that a cell type-specific change in dFB mitochondrial oxidation conveys sleep need via a redox-sensitive potassium channel, Hyperkinetic, which couples the neuron's redox state with its excitability [16]. Sleep, in turn, reverses the changes in oxidation and excitability. These results led to the hypothesis that dFB neurons monitor their own mitochondrial metabolism to gain an estimate of waking duration. To which extent they act as sentinels for oxidative hazards in-and/or outside the brain, and what drives these changes in the first place remains to be investigated. sentinels for oxidative hazards in-and/or outside the brain remains to be investigated.
Despite major advances towards a mechanistic understanding of homeostatic sleep control, many open questions still remain. What makes dFB neurons unique in their ability to monitor their own OXPHOS metabolism and to transduce mitochondrial oxidation into excitability changes? One possibility is that wakefulness sustains a dFB-specific mitochondrial mode of action that favors mtROS production, which might lead to a cumulative increase in oxidation levels over the course of waking time. Intriguingly, dFB neurons are among the cell types displaying the highest OXPHOS gene expression in the fly brain [44]. Because mtROS molecules are obligate by-products of mitochondrial aerobic respiration, it is commonly thought that mtROS production scales up as a function of the metabolic rate, i.e., when the ATP demand/consumption is high. However, somewhat counterintuitively, biochemical studies suggest the opposite scenario [45]: mtROS production is negligible when the ATP demand is high (mode 1). On the contrary, mtROS production is found to be high when the ATP demand is low and the nicotinamide adenine dinucleotide (NADH) NADH/NAD þ ratio is high (mode 2) or the ubiquinone pool (coenzyme Q) is over-reduced (mode 3, reverse electron transport) ( Figure 2). Electrophysiological recordings of dFB neurons in awake, fed flies revealed that at least some of them tend to be electrically silent. Assuming that they receive few energetically costly excitatory inputs during wakefulness, their ATP consumption and therefore demand is likely to be low, and their mitochondria prone to enter a metabolic state favoring mtROS production (modes 2 or 3). Indeed, manipulations that decrease reverse electron transport and therefore mtROS in dFB neurons lower their excitability and render flies insomniac [16], suggesting that dFB neurons operate in mode 2 and/or 3 during wakefulness. Conversely, prolonged periods of dFB activity and therefore high ATP consumption, i.e., such as during sleep, might reverse the redox changes and decrease excitability. Importantly, the existence of different mitochondrial modes of operations in neurons of behaving animals has not yet been demonstrated in vivo and is solely deduced from in vitro studies (see BOX 1). Finally, other non-mutually exclusive possibilities for a cumulative increase in mtROS are, e.g., that upstream signaling pathways trigger mtROS production, that the mitochondrial respiratory chain of dFB neurons is particularly inefficient due to a loose assembly of respiratory chain complexes, that the cellular anti-oxidant machinery is particularly poor, or that glial cells release ROS.
Given the widespread expression of Hyperkinetic in neurons, it is plausible that other non-identified Hyperkinetic-positive cells also are capable of sleep need sensing, and may use a similar redox-based mechanism. Thus, dFB neurons may not be unique in their ability to track the cellular redox state. Instead, the monitoring and tracking of OXPHOS as well as the switching between different modes of mitochondrial operations might be at the heart of an ingenious mitochondrial timing mechanism. In a broader sense, one could imagine that mitochondrial timers could operate in a variety of different cell types and circuits expressing the required components, in order to measure the build-up and discharge of a given variable, such as mtROS, over time. Testing this idea will require the simultaneous monitoring of neuronal activity and mitochondrial Modes of mitochondrial ROS production. a. In mode 1, ATP demand and respiration are high. Superoxide (O 2 -) production occurs at complex I and III but is far lower than in modes 2 and 3. b. In mode 2, the ATP demand is low and the cofactor NADH is reduced. This leads to superoxide (O 2 -) production at complex I. c. In mode 3, the ATP demand is low and the ubiquinone (Q) pool highly reduced. This leads to superoxide (O 2 -) production at complex I via reverse electron transport. The proteins of the electron transport chain are depicted in purple (I, II and III: complex I, II and III, respectively.) Q: ubiquinone; c: cytochrome c; windmill: ATPase. Electrons are represented by filled circles in white. The arrows refer to the direction of electron transport. Please note that these modes are entirely derived from in vitro studies. Modified and adapted from Refs. [1,2]. metabolism over different time scales and in different neural circuits.
Remarkably, the infusion of low levels of oxidizing substances (i.e., concentrations that do not cause oxidative damage) into the third ventricle promotes sleep in rats. In addition, microinjection of oxidizing substances into the preoptic anterior hypothalamus (POAH) elevated the sleep-inducible neuromodulators nitric oxide and adenosine, suggesting a mechanism by which peroxide may induce sleep, and in particular NREM sleep [46]. Interestingly, lesions, stimulation experiments as well electrophysiological recordings of POAH neurons provide evidence for the POAH as an important sleep-regulatory center [47e51]. This suggests that the molecular mechanism of sleep need sensing and commensurate sleep recovery might be conserved across species. However, while a cellular basis for mtROS-induced sleep has been established in Drosophila, the question of whether the POAH or other brain regions in vertebrates use similar mechanisms remains unanswered. Moreover, mammalian sleep is not uniform and is divided into distinct REM and NREM sleep stages, which may have different functions. Further studies in mammals are needed to determine whether the activity of REM-or NREMspecific neuronal populations is modulated by mtROS, and if so, whether their response is linked to a particular metabolic function in or outside the brain.

Conclusion and outlook
ROS are increasingly emerging as important signaling molecules for the homeostatic regulation of several behaviours including sleep. Future work shall uncover the mechanisms that lead to a cumulative increase of mtROS in dFB neurons as well as those that give rise to metabolic cell type specificity. Whether or not the function of sleep is to mitigate oxidative stress in the brain and/or periphery remains elusive. Until now, methodological inconsistencies and the lack of appropriate tools have led to contradictory evidence on whether sleep deprivation induces oxidative damage in the brain. Novel approaches including single-cell sequencing, lipidomics, spatial transcriptomics, advanced mass spectrometry, and new-generation ROS probes will provide critical insights into how distinct cell types in-and outside the brain respond to increased wakefulness. These approaches will likely also reveal whether the homeostatic regulation of sleep by mtROS is a conserved mechanism, both across the animal kingdom and across different brain regions.
There are several points of investigation which we have not explored in detail in this review but believe are worthy of further investigation. Firstly, given the important role of astrocytic mtROS in antioxidant neuronal protection, it will be interesting to explore the role of astrocytic mtROS in sleep regulation. Secondly, mitochondrial structure, autophagy, and biogenesis are known to be regulated by mtROS and therefore likely to be regulated by sleep-wake dynamics. They, themselves, might also regulate sleep. Finally, it will be interesting to investigate how peripheral processes affect sleep need sensing in the central brain.
In summary, mtROS are integral to both metabolism and sleep. Understanding the role of mitochondria in regulating sleep will provide critical leads into the molecular underpinnings of sleep disorders and open new approaches for the development of treatments based on nutrition and/or pharmacology. Finally, the notion that mitochondrial metabolism provides an instructive and not merely passive role in the transmission of information in the brain will provide novel, exciting avenues for neuroscience research in the future.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
No data was used for the research described in the article.